1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * NVMe over Fabrics common host code. 4 * Copyright (c) 2015-2016 HGST, a Western Digital Company. 5 */ 6 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 7 #include <linux/init.h> 8 #include <linux/miscdevice.h> 9 #include <linux/module.h> 10 #include <linux/mutex.h> 11 #include <linux/parser.h> 12 #include <linux/seq_file.h> 13 #include "nvme.h" 14 #include "fabrics.h" 15 16 static LIST_HEAD(nvmf_transports); 17 static DECLARE_RWSEM(nvmf_transports_rwsem); 18 19 static LIST_HEAD(nvmf_hosts); 20 static DEFINE_MUTEX(nvmf_hosts_mutex); 21 22 static struct nvmf_host *nvmf_default_host; 23 24 static struct nvmf_host *__nvmf_host_find(const char *hostnqn) 25 { 26 struct nvmf_host *host; 27 28 list_for_each_entry(host, &nvmf_hosts, list) { 29 if (!strcmp(host->nqn, hostnqn)) 30 return host; 31 } 32 33 return NULL; 34 } 35 36 static struct nvmf_host *nvmf_host_add(const char *hostnqn) 37 { 38 struct nvmf_host *host; 39 40 mutex_lock(&nvmf_hosts_mutex); 41 host = __nvmf_host_find(hostnqn); 42 if (host) { 43 kref_get(&host->ref); 44 goto out_unlock; 45 } 46 47 host = kmalloc(sizeof(*host), GFP_KERNEL); 48 if (!host) 49 goto out_unlock; 50 51 kref_init(&host->ref); 52 strscpy(host->nqn, hostnqn, NVMF_NQN_SIZE); 53 54 list_add_tail(&host->list, &nvmf_hosts); 55 out_unlock: 56 mutex_unlock(&nvmf_hosts_mutex); 57 return host; 58 } 59 60 static struct nvmf_host *nvmf_host_default(void) 61 { 62 struct nvmf_host *host; 63 64 host = kmalloc(sizeof(*host), GFP_KERNEL); 65 if (!host) 66 return NULL; 67 68 kref_init(&host->ref); 69 uuid_gen(&host->id); 70 snprintf(host->nqn, NVMF_NQN_SIZE, 71 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id); 72 73 mutex_lock(&nvmf_hosts_mutex); 74 list_add_tail(&host->list, &nvmf_hosts); 75 mutex_unlock(&nvmf_hosts_mutex); 76 77 return host; 78 } 79 80 static void nvmf_host_destroy(struct kref *ref) 81 { 82 struct nvmf_host *host = container_of(ref, struct nvmf_host, ref); 83 84 mutex_lock(&nvmf_hosts_mutex); 85 list_del(&host->list); 86 mutex_unlock(&nvmf_hosts_mutex); 87 88 kfree(host); 89 } 90 91 static void nvmf_host_put(struct nvmf_host *host) 92 { 93 if (host) 94 kref_put(&host->ref, nvmf_host_destroy); 95 } 96 97 /** 98 * nvmf_get_address() - Get address/port 99 * @ctrl: Host NVMe controller instance which we got the address 100 * @buf: OUTPUT parameter that will contain the address/port 101 * @size: buffer size 102 */ 103 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size) 104 { 105 int len = 0; 106 107 if (ctrl->opts->mask & NVMF_OPT_TRADDR) 108 len += scnprintf(buf, size, "traddr=%s", ctrl->opts->traddr); 109 if (ctrl->opts->mask & NVMF_OPT_TRSVCID) 110 len += scnprintf(buf + len, size - len, "%strsvcid=%s", 111 (len) ? "," : "", ctrl->opts->trsvcid); 112 if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR) 113 len += scnprintf(buf + len, size - len, "%shost_traddr=%s", 114 (len) ? "," : "", ctrl->opts->host_traddr); 115 if (ctrl->opts->mask & NVMF_OPT_HOST_IFACE) 116 len += scnprintf(buf + len, size - len, "%shost_iface=%s", 117 (len) ? "," : "", ctrl->opts->host_iface); 118 len += scnprintf(buf + len, size - len, "\n"); 119 120 return len; 121 } 122 EXPORT_SYMBOL_GPL(nvmf_get_address); 123 124 /** 125 * nvmf_reg_read32() - NVMe Fabrics "Property Get" API function. 126 * @ctrl: Host NVMe controller instance maintaining the admin 127 * queue used to submit the property read command to 128 * the allocated NVMe controller resource on the target system. 129 * @off: Starting offset value of the targeted property 130 * register (see the fabrics section of the NVMe standard). 131 * @val: OUTPUT parameter that will contain the value of 132 * the property after a successful read. 133 * 134 * Used by the host system to retrieve a 32-bit capsule property value 135 * from an NVMe controller on the target system. 136 * 137 * ("Capsule property" is an "PCIe register concept" applied to the 138 * NVMe fabrics space.) 139 * 140 * Return: 141 * 0: successful read 142 * > 0: NVMe error status code 143 * < 0: Linux errno error code 144 */ 145 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) 146 { 147 struct nvme_command cmd = { }; 148 union nvme_result res; 149 int ret; 150 151 cmd.prop_get.opcode = nvme_fabrics_command; 152 cmd.prop_get.fctype = nvme_fabrics_type_property_get; 153 cmd.prop_get.offset = cpu_to_le32(off); 154 155 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 156 NVME_QID_ANY, 0, 0); 157 158 if (ret >= 0) 159 *val = le64_to_cpu(res.u64); 160 if (unlikely(ret != 0)) 161 dev_err(ctrl->device, 162 "Property Get error: %d, offset %#x\n", 163 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 164 165 return ret; 166 } 167 EXPORT_SYMBOL_GPL(nvmf_reg_read32); 168 169 /** 170 * nvmf_reg_read64() - NVMe Fabrics "Property Get" API function. 171 * @ctrl: Host NVMe controller instance maintaining the admin 172 * queue used to submit the property read command to 173 * the allocated controller resource on the target system. 174 * @off: Starting offset value of the targeted property 175 * register (see the fabrics section of the NVMe standard). 176 * @val: OUTPUT parameter that will contain the value of 177 * the property after a successful read. 178 * 179 * Used by the host system to retrieve a 64-bit capsule property value 180 * from an NVMe controller on the target system. 181 * 182 * ("Capsule property" is an "PCIe register concept" applied to the 183 * NVMe fabrics space.) 184 * 185 * Return: 186 * 0: successful read 187 * > 0: NVMe error status code 188 * < 0: Linux errno error code 189 */ 190 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) 191 { 192 struct nvme_command cmd = { }; 193 union nvme_result res; 194 int ret; 195 196 cmd.prop_get.opcode = nvme_fabrics_command; 197 cmd.prop_get.fctype = nvme_fabrics_type_property_get; 198 cmd.prop_get.attrib = 1; 199 cmd.prop_get.offset = cpu_to_le32(off); 200 201 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, NULL, 0, 202 NVME_QID_ANY, 0, 0); 203 204 if (ret >= 0) 205 *val = le64_to_cpu(res.u64); 206 if (unlikely(ret != 0)) 207 dev_err(ctrl->device, 208 "Property Get error: %d, offset %#x\n", 209 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 210 return ret; 211 } 212 EXPORT_SYMBOL_GPL(nvmf_reg_read64); 213 214 /** 215 * nvmf_reg_write32() - NVMe Fabrics "Property Write" API function. 216 * @ctrl: Host NVMe controller instance maintaining the admin 217 * queue used to submit the property read command to 218 * the allocated NVMe controller resource on the target system. 219 * @off: Starting offset value of the targeted property 220 * register (see the fabrics section of the NVMe standard). 221 * @val: Input parameter that contains the value to be 222 * written to the property. 223 * 224 * Used by the NVMe host system to write a 32-bit capsule property value 225 * to an NVMe controller on the target system. 226 * 227 * ("Capsule property" is an "PCIe register concept" applied to the 228 * NVMe fabrics space.) 229 * 230 * Return: 231 * 0: successful write 232 * > 0: NVMe error status code 233 * < 0: Linux errno error code 234 */ 235 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) 236 { 237 struct nvme_command cmd = { }; 238 int ret; 239 240 cmd.prop_set.opcode = nvme_fabrics_command; 241 cmd.prop_set.fctype = nvme_fabrics_type_property_set; 242 cmd.prop_set.attrib = 0; 243 cmd.prop_set.offset = cpu_to_le32(off); 244 cmd.prop_set.value = cpu_to_le64(val); 245 246 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, NULL, NULL, 0, 247 NVME_QID_ANY, 0, 0); 248 if (unlikely(ret)) 249 dev_err(ctrl->device, 250 "Property Set error: %d, offset %#x\n", 251 ret > 0 ? ret & ~NVME_SC_DNR : ret, off); 252 return ret; 253 } 254 EXPORT_SYMBOL_GPL(nvmf_reg_write32); 255 256 /** 257 * nvmf_log_connect_error() - Error-parsing-diagnostic print out function for 258 * connect() errors. 259 * @ctrl: The specific /dev/nvmeX device that had the error. 260 * @errval: Error code to be decoded in a more human-friendly 261 * printout. 262 * @offset: For use with the NVMe error code 263 * NVME_SC_CONNECT_INVALID_PARAM. 264 * @cmd: This is the SQE portion of a submission capsule. 265 * @data: This is the "Data" portion of a submission capsule. 266 */ 267 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl, 268 int errval, int offset, struct nvme_command *cmd, 269 struct nvmf_connect_data *data) 270 { 271 int err_sctype = errval & ~NVME_SC_DNR; 272 273 if (errval < 0) { 274 dev_err(ctrl->device, 275 "Connect command failed, errno: %d\n", errval); 276 return; 277 } 278 279 switch (err_sctype) { 280 case NVME_SC_CONNECT_INVALID_PARAM: 281 if (offset >> 16) { 282 char *inv_data = "Connect Invalid Data Parameter"; 283 284 switch (offset & 0xffff) { 285 case (offsetof(struct nvmf_connect_data, cntlid)): 286 dev_err(ctrl->device, 287 "%s, cntlid: %d\n", 288 inv_data, data->cntlid); 289 break; 290 case (offsetof(struct nvmf_connect_data, hostnqn)): 291 dev_err(ctrl->device, 292 "%s, hostnqn \"%s\"\n", 293 inv_data, data->hostnqn); 294 break; 295 case (offsetof(struct nvmf_connect_data, subsysnqn)): 296 dev_err(ctrl->device, 297 "%s, subsysnqn \"%s\"\n", 298 inv_data, data->subsysnqn); 299 break; 300 default: 301 dev_err(ctrl->device, 302 "%s, starting byte offset: %d\n", 303 inv_data, offset & 0xffff); 304 break; 305 } 306 } else { 307 char *inv_sqe = "Connect Invalid SQE Parameter"; 308 309 switch (offset) { 310 case (offsetof(struct nvmf_connect_command, qid)): 311 dev_err(ctrl->device, 312 "%s, qid %d\n", 313 inv_sqe, cmd->connect.qid); 314 break; 315 default: 316 dev_err(ctrl->device, 317 "%s, starting byte offset: %d\n", 318 inv_sqe, offset); 319 } 320 } 321 break; 322 case NVME_SC_CONNECT_INVALID_HOST: 323 dev_err(ctrl->device, 324 "Connect for subsystem %s is not allowed, hostnqn: %s\n", 325 data->subsysnqn, data->hostnqn); 326 break; 327 case NVME_SC_CONNECT_CTRL_BUSY: 328 dev_err(ctrl->device, 329 "Connect command failed: controller is busy or not available\n"); 330 break; 331 case NVME_SC_CONNECT_FORMAT: 332 dev_err(ctrl->device, 333 "Connect incompatible format: %d", 334 cmd->connect.recfmt); 335 break; 336 case NVME_SC_HOST_PATH_ERROR: 337 dev_err(ctrl->device, 338 "Connect command failed: host path error\n"); 339 break; 340 case NVME_SC_AUTH_REQUIRED: 341 dev_err(ctrl->device, 342 "Connect command failed: authentication required\n"); 343 break; 344 default: 345 dev_err(ctrl->device, 346 "Connect command failed, error wo/DNR bit: %d\n", 347 err_sctype); 348 break; 349 } 350 } 351 352 /** 353 * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect" 354 * API function. 355 * @ctrl: Host nvme controller instance used to request 356 * a new NVMe controller allocation on the target 357 * system and establish an NVMe Admin connection to 358 * that controller. 359 * 360 * This function enables an NVMe host device to request a new allocation of 361 * an NVMe controller resource on a target system as well establish a 362 * fabrics-protocol connection of the NVMe Admin queue between the 363 * host system device and the allocated NVMe controller on the 364 * target system via a NVMe Fabrics "Connect" command. 365 * 366 * Return: 367 * 0: success 368 * > 0: NVMe error status code 369 * < 0: Linux errno error code 370 * 371 */ 372 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl) 373 { 374 struct nvme_command cmd = { }; 375 union nvme_result res; 376 struct nvmf_connect_data *data; 377 int ret; 378 u32 result; 379 380 cmd.connect.opcode = nvme_fabrics_command; 381 cmd.connect.fctype = nvme_fabrics_type_connect; 382 cmd.connect.qid = 0; 383 cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1); 384 385 /* 386 * Set keep-alive timeout in seconds granularity (ms * 1000) 387 */ 388 cmd.connect.kato = cpu_to_le32(ctrl->kato * 1000); 389 390 if (ctrl->opts->disable_sqflow) 391 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW; 392 393 data = kzalloc(sizeof(*data), GFP_KERNEL); 394 if (!data) 395 return -ENOMEM; 396 397 uuid_copy(&data->hostid, &ctrl->opts->host->id); 398 data->cntlid = cpu_to_le16(0xffff); 399 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 400 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 401 402 ret = __nvme_submit_sync_cmd(ctrl->fabrics_q, &cmd, &res, 403 data, sizeof(*data), NVME_QID_ANY, 1, 404 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT); 405 if (ret) { 406 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 407 &cmd, data); 408 goto out_free_data; 409 } 410 411 result = le32_to_cpu(res.u32); 412 ctrl->cntlid = result & 0xFFFF; 413 if ((result >> 16) & 0x3) { 414 /* Authentication required */ 415 ret = nvme_auth_negotiate(ctrl, 0); 416 if (ret) { 417 dev_warn(ctrl->device, 418 "qid 0: authentication setup failed\n"); 419 ret = NVME_SC_AUTH_REQUIRED; 420 goto out_free_data; 421 } 422 ret = nvme_auth_wait(ctrl, 0); 423 if (ret) 424 dev_warn(ctrl->device, 425 "qid 0: authentication failed\n"); 426 else 427 dev_info(ctrl->device, 428 "qid 0: authenticated\n"); 429 } 430 out_free_data: 431 kfree(data); 432 return ret; 433 } 434 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue); 435 436 /** 437 * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect" 438 * API function. 439 * @ctrl: Host nvme controller instance used to establish an 440 * NVMe I/O queue connection to the already allocated NVMe 441 * controller on the target system. 442 * @qid: NVMe I/O queue number for the new I/O connection between 443 * host and target (note qid == 0 is illegal as this is 444 * the Admin queue, per NVMe standard). 445 * 446 * This function issues a fabrics-protocol connection 447 * of a NVMe I/O queue (via NVMe Fabrics "Connect" command) 448 * between the host system device and the allocated NVMe controller 449 * on the target system. 450 * 451 * Return: 452 * 0: success 453 * > 0: NVMe error status code 454 * < 0: Linux errno error code 455 */ 456 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid) 457 { 458 struct nvme_command cmd = { }; 459 struct nvmf_connect_data *data; 460 union nvme_result res; 461 int ret; 462 u32 result; 463 464 cmd.connect.opcode = nvme_fabrics_command; 465 cmd.connect.fctype = nvme_fabrics_type_connect; 466 cmd.connect.qid = cpu_to_le16(qid); 467 cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize); 468 469 if (ctrl->opts->disable_sqflow) 470 cmd.connect.cattr |= NVME_CONNECT_DISABLE_SQFLOW; 471 472 data = kzalloc(sizeof(*data), GFP_KERNEL); 473 if (!data) 474 return -ENOMEM; 475 476 uuid_copy(&data->hostid, &ctrl->opts->host->id); 477 data->cntlid = cpu_to_le16(ctrl->cntlid); 478 strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE); 479 strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE); 480 481 ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res, 482 data, sizeof(*data), qid, 1, 483 BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT); 484 if (ret) { 485 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32), 486 &cmd, data); 487 } 488 result = le32_to_cpu(res.u32); 489 if ((result >> 16) & 2) { 490 /* Authentication required */ 491 ret = nvme_auth_negotiate(ctrl, qid); 492 if (ret) { 493 dev_warn(ctrl->device, 494 "qid %d: authentication setup failed\n", qid); 495 ret = NVME_SC_AUTH_REQUIRED; 496 } else { 497 ret = nvme_auth_wait(ctrl, qid); 498 if (ret) 499 dev_warn(ctrl->device, 500 "qid %u: authentication failed\n", qid); 501 } 502 } 503 kfree(data); 504 return ret; 505 } 506 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue); 507 508 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl) 509 { 510 if (ctrl->opts->max_reconnects == -1 || 511 ctrl->nr_reconnects < ctrl->opts->max_reconnects) 512 return true; 513 514 return false; 515 } 516 EXPORT_SYMBOL_GPL(nvmf_should_reconnect); 517 518 /** 519 * nvmf_register_transport() - NVMe Fabrics Library registration function. 520 * @ops: Transport ops instance to be registered to the 521 * common fabrics library. 522 * 523 * API function that registers the type of specific transport fabric 524 * being implemented to the common NVMe fabrics library. Part of 525 * the overall init sequence of starting up a fabrics driver. 526 */ 527 int nvmf_register_transport(struct nvmf_transport_ops *ops) 528 { 529 if (!ops->create_ctrl) 530 return -EINVAL; 531 532 down_write(&nvmf_transports_rwsem); 533 list_add_tail(&ops->entry, &nvmf_transports); 534 up_write(&nvmf_transports_rwsem); 535 536 return 0; 537 } 538 EXPORT_SYMBOL_GPL(nvmf_register_transport); 539 540 /** 541 * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function. 542 * @ops: Transport ops instance to be unregistered from the 543 * common fabrics library. 544 * 545 * Fabrics API function that unregisters the type of specific transport 546 * fabric being implemented from the common NVMe fabrics library. 547 * Part of the overall exit sequence of unloading the implemented driver. 548 */ 549 void nvmf_unregister_transport(struct nvmf_transport_ops *ops) 550 { 551 down_write(&nvmf_transports_rwsem); 552 list_del(&ops->entry); 553 up_write(&nvmf_transports_rwsem); 554 } 555 EXPORT_SYMBOL_GPL(nvmf_unregister_transport); 556 557 static struct nvmf_transport_ops *nvmf_lookup_transport( 558 struct nvmf_ctrl_options *opts) 559 { 560 struct nvmf_transport_ops *ops; 561 562 lockdep_assert_held(&nvmf_transports_rwsem); 563 564 list_for_each_entry(ops, &nvmf_transports, entry) { 565 if (strcmp(ops->name, opts->transport) == 0) 566 return ops; 567 } 568 569 return NULL; 570 } 571 572 static const match_table_t opt_tokens = { 573 { NVMF_OPT_TRANSPORT, "transport=%s" }, 574 { NVMF_OPT_TRADDR, "traddr=%s" }, 575 { NVMF_OPT_TRSVCID, "trsvcid=%s" }, 576 { NVMF_OPT_NQN, "nqn=%s" }, 577 { NVMF_OPT_QUEUE_SIZE, "queue_size=%d" }, 578 { NVMF_OPT_NR_IO_QUEUES, "nr_io_queues=%d" }, 579 { NVMF_OPT_RECONNECT_DELAY, "reconnect_delay=%d" }, 580 { NVMF_OPT_CTRL_LOSS_TMO, "ctrl_loss_tmo=%d" }, 581 { NVMF_OPT_KATO, "keep_alive_tmo=%d" }, 582 { NVMF_OPT_HOSTNQN, "hostnqn=%s" }, 583 { NVMF_OPT_HOST_TRADDR, "host_traddr=%s" }, 584 { NVMF_OPT_HOST_IFACE, "host_iface=%s" }, 585 { NVMF_OPT_HOST_ID, "hostid=%s" }, 586 { NVMF_OPT_DUP_CONNECT, "duplicate_connect" }, 587 { NVMF_OPT_DISABLE_SQFLOW, "disable_sqflow" }, 588 { NVMF_OPT_HDR_DIGEST, "hdr_digest" }, 589 { NVMF_OPT_DATA_DIGEST, "data_digest" }, 590 { NVMF_OPT_NR_WRITE_QUEUES, "nr_write_queues=%d" }, 591 { NVMF_OPT_NR_POLL_QUEUES, "nr_poll_queues=%d" }, 592 { NVMF_OPT_TOS, "tos=%d" }, 593 { NVMF_OPT_FAIL_FAST_TMO, "fast_io_fail_tmo=%d" }, 594 { NVMF_OPT_DISCOVERY, "discovery" }, 595 { NVMF_OPT_DHCHAP_SECRET, "dhchap_secret=%s" }, 596 { NVMF_OPT_DHCHAP_CTRL_SECRET, "dhchap_ctrl_secret=%s" }, 597 { NVMF_OPT_ERR, NULL } 598 }; 599 600 static int nvmf_parse_options(struct nvmf_ctrl_options *opts, 601 const char *buf) 602 { 603 substring_t args[MAX_OPT_ARGS]; 604 char *options, *o, *p; 605 int token, ret = 0; 606 size_t nqnlen = 0; 607 int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO; 608 uuid_t hostid; 609 610 /* Set defaults */ 611 opts->queue_size = NVMF_DEF_QUEUE_SIZE; 612 opts->nr_io_queues = num_online_cpus(); 613 opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY; 614 opts->kato = 0; 615 opts->duplicate_connect = false; 616 opts->fast_io_fail_tmo = NVMF_DEF_FAIL_FAST_TMO; 617 opts->hdr_digest = false; 618 opts->data_digest = false; 619 opts->tos = -1; /* < 0 == use transport default */ 620 621 options = o = kstrdup(buf, GFP_KERNEL); 622 if (!options) 623 return -ENOMEM; 624 625 uuid_gen(&hostid); 626 627 while ((p = strsep(&o, ",\n")) != NULL) { 628 if (!*p) 629 continue; 630 631 token = match_token(p, opt_tokens, args); 632 opts->mask |= token; 633 switch (token) { 634 case NVMF_OPT_TRANSPORT: 635 p = match_strdup(args); 636 if (!p) { 637 ret = -ENOMEM; 638 goto out; 639 } 640 kfree(opts->transport); 641 opts->transport = p; 642 break; 643 case NVMF_OPT_NQN: 644 p = match_strdup(args); 645 if (!p) { 646 ret = -ENOMEM; 647 goto out; 648 } 649 kfree(opts->subsysnqn); 650 opts->subsysnqn = p; 651 nqnlen = strlen(opts->subsysnqn); 652 if (nqnlen >= NVMF_NQN_SIZE) { 653 pr_err("%s needs to be < %d bytes\n", 654 opts->subsysnqn, NVMF_NQN_SIZE); 655 ret = -EINVAL; 656 goto out; 657 } 658 opts->discovery_nqn = 659 !(strcmp(opts->subsysnqn, 660 NVME_DISC_SUBSYS_NAME)); 661 break; 662 case NVMF_OPT_TRADDR: 663 p = match_strdup(args); 664 if (!p) { 665 ret = -ENOMEM; 666 goto out; 667 } 668 kfree(opts->traddr); 669 opts->traddr = p; 670 break; 671 case NVMF_OPT_TRSVCID: 672 p = match_strdup(args); 673 if (!p) { 674 ret = -ENOMEM; 675 goto out; 676 } 677 kfree(opts->trsvcid); 678 opts->trsvcid = p; 679 break; 680 case NVMF_OPT_QUEUE_SIZE: 681 if (match_int(args, &token)) { 682 ret = -EINVAL; 683 goto out; 684 } 685 if (token < NVMF_MIN_QUEUE_SIZE || 686 token > NVMF_MAX_QUEUE_SIZE) { 687 pr_err("Invalid queue_size %d\n", token); 688 ret = -EINVAL; 689 goto out; 690 } 691 opts->queue_size = token; 692 break; 693 case NVMF_OPT_NR_IO_QUEUES: 694 if (match_int(args, &token)) { 695 ret = -EINVAL; 696 goto out; 697 } 698 if (token <= 0) { 699 pr_err("Invalid number of IOQs %d\n", token); 700 ret = -EINVAL; 701 goto out; 702 } 703 if (opts->discovery_nqn) { 704 pr_debug("Ignoring nr_io_queues value for discovery controller\n"); 705 break; 706 } 707 708 opts->nr_io_queues = min_t(unsigned int, 709 num_online_cpus(), token); 710 break; 711 case NVMF_OPT_KATO: 712 if (match_int(args, &token)) { 713 ret = -EINVAL; 714 goto out; 715 } 716 717 if (token < 0) { 718 pr_err("Invalid keep_alive_tmo %d\n", token); 719 ret = -EINVAL; 720 goto out; 721 } else if (token == 0 && !opts->discovery_nqn) { 722 /* Allowed for debug */ 723 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n"); 724 } 725 opts->kato = token; 726 break; 727 case NVMF_OPT_CTRL_LOSS_TMO: 728 if (match_int(args, &token)) { 729 ret = -EINVAL; 730 goto out; 731 } 732 733 if (token < 0) 734 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n"); 735 ctrl_loss_tmo = token; 736 break; 737 case NVMF_OPT_FAIL_FAST_TMO: 738 if (match_int(args, &token)) { 739 ret = -EINVAL; 740 goto out; 741 } 742 743 if (token >= 0) 744 pr_warn("I/O fail on reconnect controller after %d sec\n", 745 token); 746 else 747 token = -1; 748 749 opts->fast_io_fail_tmo = token; 750 break; 751 case NVMF_OPT_HOSTNQN: 752 if (opts->host) { 753 pr_err("hostnqn already user-assigned: %s\n", 754 opts->host->nqn); 755 ret = -EADDRINUSE; 756 goto out; 757 } 758 p = match_strdup(args); 759 if (!p) { 760 ret = -ENOMEM; 761 goto out; 762 } 763 nqnlen = strlen(p); 764 if (nqnlen >= NVMF_NQN_SIZE) { 765 pr_err("%s needs to be < %d bytes\n", 766 p, NVMF_NQN_SIZE); 767 kfree(p); 768 ret = -EINVAL; 769 goto out; 770 } 771 opts->host = nvmf_host_add(p); 772 kfree(p); 773 if (!opts->host) { 774 ret = -ENOMEM; 775 goto out; 776 } 777 break; 778 case NVMF_OPT_RECONNECT_DELAY: 779 if (match_int(args, &token)) { 780 ret = -EINVAL; 781 goto out; 782 } 783 if (token <= 0) { 784 pr_err("Invalid reconnect_delay %d\n", token); 785 ret = -EINVAL; 786 goto out; 787 } 788 opts->reconnect_delay = token; 789 break; 790 case NVMF_OPT_HOST_TRADDR: 791 p = match_strdup(args); 792 if (!p) { 793 ret = -ENOMEM; 794 goto out; 795 } 796 kfree(opts->host_traddr); 797 opts->host_traddr = p; 798 break; 799 case NVMF_OPT_HOST_IFACE: 800 p = match_strdup(args); 801 if (!p) { 802 ret = -ENOMEM; 803 goto out; 804 } 805 kfree(opts->host_iface); 806 opts->host_iface = p; 807 break; 808 case NVMF_OPT_HOST_ID: 809 p = match_strdup(args); 810 if (!p) { 811 ret = -ENOMEM; 812 goto out; 813 } 814 ret = uuid_parse(p, &hostid); 815 if (ret) { 816 pr_err("Invalid hostid %s\n", p); 817 ret = -EINVAL; 818 kfree(p); 819 goto out; 820 } 821 kfree(p); 822 break; 823 case NVMF_OPT_DUP_CONNECT: 824 opts->duplicate_connect = true; 825 break; 826 case NVMF_OPT_DISABLE_SQFLOW: 827 opts->disable_sqflow = true; 828 break; 829 case NVMF_OPT_HDR_DIGEST: 830 opts->hdr_digest = true; 831 break; 832 case NVMF_OPT_DATA_DIGEST: 833 opts->data_digest = true; 834 break; 835 case NVMF_OPT_NR_WRITE_QUEUES: 836 if (match_int(args, &token)) { 837 ret = -EINVAL; 838 goto out; 839 } 840 if (token <= 0) { 841 pr_err("Invalid nr_write_queues %d\n", token); 842 ret = -EINVAL; 843 goto out; 844 } 845 opts->nr_write_queues = token; 846 break; 847 case NVMF_OPT_NR_POLL_QUEUES: 848 if (match_int(args, &token)) { 849 ret = -EINVAL; 850 goto out; 851 } 852 if (token <= 0) { 853 pr_err("Invalid nr_poll_queues %d\n", token); 854 ret = -EINVAL; 855 goto out; 856 } 857 opts->nr_poll_queues = token; 858 break; 859 case NVMF_OPT_TOS: 860 if (match_int(args, &token)) { 861 ret = -EINVAL; 862 goto out; 863 } 864 if (token < 0) { 865 pr_err("Invalid type of service %d\n", token); 866 ret = -EINVAL; 867 goto out; 868 } 869 if (token > 255) { 870 pr_warn("Clamping type of service to 255\n"); 871 token = 255; 872 } 873 opts->tos = token; 874 break; 875 case NVMF_OPT_DISCOVERY: 876 opts->discovery_nqn = true; 877 break; 878 case NVMF_OPT_DHCHAP_SECRET: 879 p = match_strdup(args); 880 if (!p) { 881 ret = -ENOMEM; 882 goto out; 883 } 884 if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) { 885 pr_err("Invalid DH-CHAP secret %s\n", p); 886 ret = -EINVAL; 887 goto out; 888 } 889 kfree(opts->dhchap_secret); 890 opts->dhchap_secret = p; 891 break; 892 case NVMF_OPT_DHCHAP_CTRL_SECRET: 893 p = match_strdup(args); 894 if (!p) { 895 ret = -ENOMEM; 896 goto out; 897 } 898 if (strlen(p) < 11 || strncmp(p, "DHHC-1:", 7)) { 899 pr_err("Invalid DH-CHAP secret %s\n", p); 900 ret = -EINVAL; 901 goto out; 902 } 903 kfree(opts->dhchap_ctrl_secret); 904 opts->dhchap_ctrl_secret = p; 905 break; 906 default: 907 pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n", 908 p); 909 ret = -EINVAL; 910 goto out; 911 } 912 } 913 914 if (opts->discovery_nqn) { 915 opts->nr_io_queues = 0; 916 opts->nr_write_queues = 0; 917 opts->nr_poll_queues = 0; 918 opts->duplicate_connect = true; 919 } else { 920 if (!opts->kato) 921 opts->kato = NVME_DEFAULT_KATO; 922 } 923 if (ctrl_loss_tmo < 0) { 924 opts->max_reconnects = -1; 925 } else { 926 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo, 927 opts->reconnect_delay); 928 if (ctrl_loss_tmo < opts->fast_io_fail_tmo) 929 pr_warn("failfast tmo (%d) larger than controller loss tmo (%d)\n", 930 opts->fast_io_fail_tmo, ctrl_loss_tmo); 931 } 932 933 if (!opts->host) { 934 kref_get(&nvmf_default_host->ref); 935 opts->host = nvmf_default_host; 936 } 937 938 uuid_copy(&opts->host->id, &hostid); 939 940 out: 941 kfree(options); 942 return ret; 943 } 944 945 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts, 946 unsigned int required_opts) 947 { 948 if ((opts->mask & required_opts) != required_opts) { 949 unsigned int i; 950 951 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 952 if ((opt_tokens[i].token & required_opts) && 953 !(opt_tokens[i].token & opts->mask)) { 954 pr_warn("missing parameter '%s'\n", 955 opt_tokens[i].pattern); 956 } 957 } 958 959 return -EINVAL; 960 } 961 962 return 0; 963 } 964 965 bool nvmf_ip_options_match(struct nvme_ctrl *ctrl, 966 struct nvmf_ctrl_options *opts) 967 { 968 if (!nvmf_ctlr_matches_baseopts(ctrl, opts) || 969 strcmp(opts->traddr, ctrl->opts->traddr) || 970 strcmp(opts->trsvcid, ctrl->opts->trsvcid)) 971 return false; 972 973 /* 974 * Checking the local address or host interfaces is rough. 975 * 976 * In most cases, none is specified and the host port or 977 * host interface is selected by the stack. 978 * 979 * Assume no match if: 980 * - local address or host interface is specified and address 981 * or host interface is not the same 982 * - local address or host interface is not specified but 983 * remote is, or vice versa (admin using specific 984 * host_traddr/host_iface when it matters). 985 */ 986 if ((opts->mask & NVMF_OPT_HOST_TRADDR) && 987 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) { 988 if (strcmp(opts->host_traddr, ctrl->opts->host_traddr)) 989 return false; 990 } else if ((opts->mask & NVMF_OPT_HOST_TRADDR) || 991 (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)) { 992 return false; 993 } 994 995 if ((opts->mask & NVMF_OPT_HOST_IFACE) && 996 (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) { 997 if (strcmp(opts->host_iface, ctrl->opts->host_iface)) 998 return false; 999 } else if ((opts->mask & NVMF_OPT_HOST_IFACE) || 1000 (ctrl->opts->mask & NVMF_OPT_HOST_IFACE)) { 1001 return false; 1002 } 1003 1004 return true; 1005 } 1006 EXPORT_SYMBOL_GPL(nvmf_ip_options_match); 1007 1008 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts, 1009 unsigned int allowed_opts) 1010 { 1011 if (opts->mask & ~allowed_opts) { 1012 unsigned int i; 1013 1014 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) { 1015 if ((opt_tokens[i].token & opts->mask) && 1016 (opt_tokens[i].token & ~allowed_opts)) { 1017 pr_warn("invalid parameter '%s'\n", 1018 opt_tokens[i].pattern); 1019 } 1020 } 1021 1022 return -EINVAL; 1023 } 1024 1025 return 0; 1026 } 1027 1028 void nvmf_free_options(struct nvmf_ctrl_options *opts) 1029 { 1030 nvmf_host_put(opts->host); 1031 kfree(opts->transport); 1032 kfree(opts->traddr); 1033 kfree(opts->trsvcid); 1034 kfree(opts->subsysnqn); 1035 kfree(opts->host_traddr); 1036 kfree(opts->host_iface); 1037 kfree(opts->dhchap_secret); 1038 kfree(opts->dhchap_ctrl_secret); 1039 kfree(opts); 1040 } 1041 EXPORT_SYMBOL_GPL(nvmf_free_options); 1042 1043 #define NVMF_REQUIRED_OPTS (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN) 1044 #define NVMF_ALLOWED_OPTS (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \ 1045 NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \ 1046 NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT |\ 1047 NVMF_OPT_DISABLE_SQFLOW | NVMF_OPT_DISCOVERY |\ 1048 NVMF_OPT_FAIL_FAST_TMO | NVMF_OPT_DHCHAP_SECRET |\ 1049 NVMF_OPT_DHCHAP_CTRL_SECRET) 1050 1051 static struct nvme_ctrl * 1052 nvmf_create_ctrl(struct device *dev, const char *buf) 1053 { 1054 struct nvmf_ctrl_options *opts; 1055 struct nvmf_transport_ops *ops; 1056 struct nvme_ctrl *ctrl; 1057 int ret; 1058 1059 opts = kzalloc(sizeof(*opts), GFP_KERNEL); 1060 if (!opts) 1061 return ERR_PTR(-ENOMEM); 1062 1063 ret = nvmf_parse_options(opts, buf); 1064 if (ret) 1065 goto out_free_opts; 1066 1067 1068 request_module("nvme-%s", opts->transport); 1069 1070 /* 1071 * Check the generic options first as we need a valid transport for 1072 * the lookup below. Then clear the generic flags so that transport 1073 * drivers don't have to care about them. 1074 */ 1075 ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS); 1076 if (ret) 1077 goto out_free_opts; 1078 opts->mask &= ~NVMF_REQUIRED_OPTS; 1079 1080 down_read(&nvmf_transports_rwsem); 1081 ops = nvmf_lookup_transport(opts); 1082 if (!ops) { 1083 pr_info("no handler found for transport %s.\n", 1084 opts->transport); 1085 ret = -EINVAL; 1086 goto out_unlock; 1087 } 1088 1089 if (!try_module_get(ops->module)) { 1090 ret = -EBUSY; 1091 goto out_unlock; 1092 } 1093 up_read(&nvmf_transports_rwsem); 1094 1095 ret = nvmf_check_required_opts(opts, ops->required_opts); 1096 if (ret) 1097 goto out_module_put; 1098 ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS | 1099 ops->allowed_opts | ops->required_opts); 1100 if (ret) 1101 goto out_module_put; 1102 1103 ctrl = ops->create_ctrl(dev, opts); 1104 if (IS_ERR(ctrl)) { 1105 ret = PTR_ERR(ctrl); 1106 goto out_module_put; 1107 } 1108 1109 module_put(ops->module); 1110 return ctrl; 1111 1112 out_module_put: 1113 module_put(ops->module); 1114 goto out_free_opts; 1115 out_unlock: 1116 up_read(&nvmf_transports_rwsem); 1117 out_free_opts: 1118 nvmf_free_options(opts); 1119 return ERR_PTR(ret); 1120 } 1121 1122 static struct class *nvmf_class; 1123 static struct device *nvmf_device; 1124 static DEFINE_MUTEX(nvmf_dev_mutex); 1125 1126 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf, 1127 size_t count, loff_t *pos) 1128 { 1129 struct seq_file *seq_file = file->private_data; 1130 struct nvme_ctrl *ctrl; 1131 const char *buf; 1132 int ret = 0; 1133 1134 if (count > PAGE_SIZE) 1135 return -ENOMEM; 1136 1137 buf = memdup_user_nul(ubuf, count); 1138 if (IS_ERR(buf)) 1139 return PTR_ERR(buf); 1140 1141 mutex_lock(&nvmf_dev_mutex); 1142 if (seq_file->private) { 1143 ret = -EINVAL; 1144 goto out_unlock; 1145 } 1146 1147 ctrl = nvmf_create_ctrl(nvmf_device, buf); 1148 if (IS_ERR(ctrl)) { 1149 ret = PTR_ERR(ctrl); 1150 goto out_unlock; 1151 } 1152 1153 seq_file->private = ctrl; 1154 1155 out_unlock: 1156 mutex_unlock(&nvmf_dev_mutex); 1157 kfree(buf); 1158 return ret ? ret : count; 1159 } 1160 1161 static void __nvmf_concat_opt_tokens(struct seq_file *seq_file) 1162 { 1163 const struct match_token *tok; 1164 int idx; 1165 1166 /* 1167 * Add dummy entries for instance and cntlid to 1168 * signal an invalid/non-existing controller 1169 */ 1170 seq_puts(seq_file, "instance=-1,cntlid=-1"); 1171 for (idx = 0; idx < ARRAY_SIZE(opt_tokens); idx++) { 1172 tok = &opt_tokens[idx]; 1173 if (tok->token == NVMF_OPT_ERR) 1174 continue; 1175 seq_puts(seq_file, ","); 1176 seq_puts(seq_file, tok->pattern); 1177 } 1178 seq_puts(seq_file, "\n"); 1179 } 1180 1181 static int nvmf_dev_show(struct seq_file *seq_file, void *private) 1182 { 1183 struct nvme_ctrl *ctrl; 1184 1185 mutex_lock(&nvmf_dev_mutex); 1186 ctrl = seq_file->private; 1187 if (!ctrl) { 1188 __nvmf_concat_opt_tokens(seq_file); 1189 goto out_unlock; 1190 } 1191 1192 seq_printf(seq_file, "instance=%d,cntlid=%d\n", 1193 ctrl->instance, ctrl->cntlid); 1194 1195 out_unlock: 1196 mutex_unlock(&nvmf_dev_mutex); 1197 return 0; 1198 } 1199 1200 static int nvmf_dev_open(struct inode *inode, struct file *file) 1201 { 1202 /* 1203 * The miscdevice code initializes file->private_data, but doesn't 1204 * make use of it later. 1205 */ 1206 file->private_data = NULL; 1207 return single_open(file, nvmf_dev_show, NULL); 1208 } 1209 1210 static int nvmf_dev_release(struct inode *inode, struct file *file) 1211 { 1212 struct seq_file *seq_file = file->private_data; 1213 struct nvme_ctrl *ctrl = seq_file->private; 1214 1215 if (ctrl) 1216 nvme_put_ctrl(ctrl); 1217 return single_release(inode, file); 1218 } 1219 1220 static const struct file_operations nvmf_dev_fops = { 1221 .owner = THIS_MODULE, 1222 .write = nvmf_dev_write, 1223 .read = seq_read, 1224 .open = nvmf_dev_open, 1225 .release = nvmf_dev_release, 1226 }; 1227 1228 static struct miscdevice nvmf_misc = { 1229 .minor = MISC_DYNAMIC_MINOR, 1230 .name = "nvme-fabrics", 1231 .fops = &nvmf_dev_fops, 1232 }; 1233 1234 static int __init nvmf_init(void) 1235 { 1236 int ret; 1237 1238 nvmf_default_host = nvmf_host_default(); 1239 if (!nvmf_default_host) 1240 return -ENOMEM; 1241 1242 nvmf_class = class_create(THIS_MODULE, "nvme-fabrics"); 1243 if (IS_ERR(nvmf_class)) { 1244 pr_err("couldn't register class nvme-fabrics\n"); 1245 ret = PTR_ERR(nvmf_class); 1246 goto out_free_host; 1247 } 1248 1249 nvmf_device = 1250 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl"); 1251 if (IS_ERR(nvmf_device)) { 1252 pr_err("couldn't create nvme-fabrics device!\n"); 1253 ret = PTR_ERR(nvmf_device); 1254 goto out_destroy_class; 1255 } 1256 1257 ret = misc_register(&nvmf_misc); 1258 if (ret) { 1259 pr_err("couldn't register misc device: %d\n", ret); 1260 goto out_destroy_device; 1261 } 1262 1263 return 0; 1264 1265 out_destroy_device: 1266 device_destroy(nvmf_class, MKDEV(0, 0)); 1267 out_destroy_class: 1268 class_destroy(nvmf_class); 1269 out_free_host: 1270 nvmf_host_put(nvmf_default_host); 1271 return ret; 1272 } 1273 1274 static void __exit nvmf_exit(void) 1275 { 1276 misc_deregister(&nvmf_misc); 1277 device_destroy(nvmf_class, MKDEV(0, 0)); 1278 class_destroy(nvmf_class); 1279 nvmf_host_put(nvmf_default_host); 1280 1281 BUILD_BUG_ON(sizeof(struct nvmf_common_command) != 64); 1282 BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64); 1283 BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64); 1284 BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64); 1285 BUILD_BUG_ON(sizeof(struct nvmf_auth_send_command) != 64); 1286 BUILD_BUG_ON(sizeof(struct nvmf_auth_receive_command) != 64); 1287 BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024); 1288 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_negotiate_data) != 8); 1289 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_challenge_data) != 16); 1290 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_reply_data) != 16); 1291 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success1_data) != 16); 1292 BUILD_BUG_ON(sizeof(struct nvmf_auth_dhchap_success2_data) != 16); 1293 } 1294 1295 MODULE_LICENSE("GPL v2"); 1296 1297 module_init(nvmf_init); 1298 module_exit(nvmf_exit); 1299